【摘要】 开发适合集群系统的并行数值算法,可以解决更多的科学与工程计算问题。在PC集群环境下,提出对Cannon算法的改进方案,采用重叠技术,使数据的计算和通信在时间上重叠,达到时延隐藏的目的,克服了网络传输的通信瓶颈问题;还提出一种自适应负载反馈平衡策略,很好地解决了集群负载平衡。该算法在PC集群系统环境下测试,取得了较好的并行效率,在4个结点上和8个结点上,加速比分别达到3.77和7.93。更多还原

【Abstract】 Developing appropriate numerical algorithm in cluster system in order to solve more science and engineering problem.Cannon algorithm is improved by making computing and communication synchro in order to hide delay-time and solve the communication bottle-neck of network visiting.An adaptive adjustment load feedback balancing policy is presented so that the system can keep the load balance.The result of test explain that parallel performance can significantly improved in PC cluster system environment.Speedup is 3.77 in 4-nodes cluster system environment,7.93 in 8-nodes cluster system environment.更多还原